Industry Research

Appearance

V2X Cooperative Planning ​

Purpose: Extend V2X from communication and cooperative perception into the planning loop: prediction, tactical maneuver choice, trajectory negotiation, reservation, fallback, and deployment assurance for connected autonomous vehicles.

Key Takeaway: V2X becomes operationally valuable when it changes what the vehicle plans to do. Status and perception sharing improve awareness, but cooperative planning requires intent, maneuver, authority, and trust semantics that can be consumed by behavior planning and trajectory generation. Airside fleets are an unusually strong fit because participants are bounded, known, low-speed, infrastructure-rich, and centrally governed.

Research current as of: 2026-05-09

Problem Framing ​

Most V2X documents stop at radios, message types, and cooperative perception. That is necessary but incomplete. A vehicle still needs to decide whether to yield, request priority, enter a single-lane segment, clear a stand, or reject stale advice. Cooperative planning is the layer that converts communicated state and intent into behavior and trajectory decisions.

For airside autonomy, V2X can supply information that onboard sensing cannot reliably infer:

  • Aircraft and GSE intent behind occlusions.
  • Stand phase, pushback, fuel, de-icing, emergency, and turnaround status.
  • Road-resource reservations for narrow service roads and stand pockets.
  • Infrastructure-detected hazards such as FOD, jet blast, personnel, and temporary closures.
  • Ramp-control or fleet-manager advice for coordinated maneuvers.

The design principle is conservative: V2X can increase capability, but loss or corruption of V2X must not remove the vehicle's ability to stop safely and obey default-deny authority rules.

Method and Architecture Taxonomy ​

Cooperation Classes ​

ETSI's Manoeuvre Coordination Service and SAE cooperative-driving terminology are useful for separating levels of cooperation:

ClassMeaningPlanning Impact
Status sharing"Here is my current state"Better tracking and TTC estimates
Intent sharing"Here is what I plan to do"Conditional prediction and earlier yielding
Agreement seeking"Can we execute this joint maneuver?"Negotiated merge, crossing, convoy, or single-lane reservation
Prescriptive"An authorized entity advises or commands this maneuver"Infrastructure or ramp-control priority and route allocation

Airside examples:

  • Status: Tug broadcasts pose, speed, vehicle type, task, and brake/turn state.
  • Intent: Pushback tug broadcasts planned aircraft sweep path and time window.
  • Agreement seeking: Two GSE negotiate entry to a one-lane service road.
  • Prescriptive: Fleet manager or ramp control assigns a vehicle to hold/clear/proceed.

Cooperative Planning Data Plane ​

Data TypeSourcePlanning Use
Cooperative awareness / statusVehicle, RSU, infrastructureTrack confirmation, TTC, right-of-way estimation
Cooperative perceptionVehicle/infrastructure sensorsOccluded object occupancy and confidence
Intent and trajectoryVehicles, aircraft-tug systems, fleet managerConditional prediction and behavior scoring
Maneuver requests/advicePeer vehicles, infrastructure, traffic managerAgreement-seeking and prescriptive coordination
Operational authorityA-CDM, A-SMGCS, ramp control, NOTAMsHold/proceed, route validity, mission priority
Road-resource reservationsFleet traffic manager or distributed protocolDeadlock prevention and narrow-zone access
Trust and healthPKI, misbehavior monitor, network QoSMessage weighting, rejection, degraded mode

Planning Loop Integration ​

text
Receive V2X / V2I / V2N messages
  -> authenticate, time-align, and geofence
  -> reject stale, inconsistent, or unauthorized messages
  -> fuse into cooperative world model
  -> condition prediction on shared intent
  -> behavior arbitration: yield, request, reserve, proceed, clear, stop
  -> trajectory generation within accepted reservations and authority state
  -> publish own status, intent, and reservation request/update
  -> monitor execution and revoke/rollback on deviation

The ego planner should never use V2X as a direct actuator command. V2X updates the world model, constraints, and tactical intent; the local planner and validator still produce and check the trajectory.

System Architectures ​

ArchitectureDescriptionAirside Fit
Local advisory V2XVehicle consumes peer status/intent as extra prediction inputGood first step, low certification burden
Central traffic managerEdge server allocates road resources and prioritiesStrong fit for airports and yards with private networks
Decentralized agreementVehicles negotiate reservations directlyUseful fallback when central link is unavailable
HybridCentral manager for strategic/tactical zones, local V2V for immediate negotiationRecommended for airside fleets
End-to-end cooperative plannerNeural model fuses ego and infrastructure data through to trajectoryResearch frontier; useful in shadow mode before deployment

End-to-End Cooperative Planning Research ​

Recent V2X E2E work shows the field moving beyond perception-only gains:

  • UniV2X integrates vehicle-infrastructure cooperation across perception, mapping, occupancy prediction, and planning with sparse-dense transmission.
  • UniE2EV2X emphasizes unified V2X cooperative driving with accident prediction and end-to-end fusion.
  • V2X-VLM combines vehicle and infrastructure camera views with text scene descriptions for cooperative trajectory planning.

For airside deployment, these are not drop-in production planners. They are useful patterns for shadow-mode scoring, cooperative feature design, and future learned cost/proposal modules. The deployable near-term stack should keep explicit authority, trust, and safety validation around any learned cooperative planner.

Evaluation and Deployment Notes ​

Metrics ​

CategoryMetrics
Planning qualityroute completion, success rate, progress, comfort, unnecessary stops
Cooperative valuedelta vs no-V2X baseline, occlusion resolution, earlier yielding, deadlock reduction
Safetycollision rate, TTC, clearance violations, hold-short compliance, jet-blast/no-go-zone entry
Communicationlatency, jitter, packet loss, bandwidth, stale-message rejection, congestion behavior
Trust and securityinvalid signature rejection, inconsistency detection, misbehavior demotion, replay resistance
Robustnessscore under partial participation, infrastructure outage, wrong intent, clock drift, localization error
Operationsblocked-zone time, reservation conflicts, remote-assistance calls, mission lateness

Always report V2X-enabled performance against at least three baselines:

  1. Onboard-only planning.
  2. V2X status/perception only, with no intent or maneuver coordination.
  3. Full cooperative planning with intent/reservation/advice.

This prevents inflated claims where V2X appears useful only because the onboard-only baseline is weak.

Deployment Stages ​

  1. Listen-only: receive messages, log what decisions would have changed, never affect control.
  2. Advisory prediction: use status/intent to adjust prediction costs but keep conservative behavior.
  3. Local constraints: consume infrastructure hazards and reservations as validated constraints.
  4. Agreement seeking: negotiate one-lane segments, stand-entry, and crossing maneuvers in controlled zones.
  5. Prescriptive coordination: accept authorized fleet/ramp-control advice with explicit TTL and fallback.

Deployment requirements:

  • All messages used for planning need source identity, timestamp, frame, covariance/confidence, TTL, and authority class.
  • Maneuver advice must be checked against local map, perception, safety constraints, and route authority.
  • Vehicle intent broadcasts should include a confidence and a revocation/update path if ego deviates.
  • Clock synchronization and timestamp provenance are safety requirements, not logging niceties.
  • The fallback policy must be deterministic for V2X loss: slow, inflate margins, hold at authority-critical boundaries, and stop when necessary.

Indoor / Outdoor / Airside Fit ​

DomainFitCooperative Planning Pattern
Indoor warehouse / factoryHighFleet manager reserves aisles, doors, lifts, charging bays; robot-to-robot status fills blind corners
Outdoor yard / depotVery highYard manager reserves gates, trailer lanes, staging areas, and one-lane roads
Public-road AVMedium to highValuable for intersections, merges, emergency vehicles, and infrastructure sensing; partial adoption is hard
Airside apronVery highKnown fleet, private 5G/CBRS, airport authority, stand sequencing, aircraft/GSE priority
Movement areaHigh with strict authorityV2X can aid awareness, but explicit clearance and default-deny rules must dominate

Airside is the best near-term domain for cooperative planning because the trust domain is closed and the infrastructure owner can require participation. The main caveat is mixed traffic: human-driven GSE, personnel, and aircraft may remain unconnected for years, so onboard sensing and conservative rules must remain authoritative.

Failure Modes ​

Failure ModeSymptomMitigation
Stale V2X intentVehicle yields to a maneuver that is no longer happeningTTL, monotonic sequence numbers, execution monitoring
False cooperative objectPlanner avoids a nonexistent or spoofed vehiclePKI, plausibility checks, onboard perception cross-check
Missing unconnected actorV2X-planned reservation ignores a pedestrian or legacy GSENever treat V2X as complete occupancy; fuse with onboard/infrastructure perception
Conflicting advicePeer, fleet manager, and local rules disagreeAuthority hierarchy, rule engine, default safe stop
Clock driftTrajectories are time-shifted and collidegPTP/PTP monitoring, timestamp uncertainty inflation
Localization frame mismatchShared trajectories are spatially offsetframe IDs, map-version checks, calibration monitors
Packet loss / congestionNegotiation fails or intent updates are delayeddeterministic fallback, DCC/geofencing, bounded message rates
OvercentralizationFleet manager outage blocks local mobilityhybrid fallback with local V2V and preapproved safe behaviors
Overtrust in learned E2E V2XNeural planner exploits cooperative signals without explainable constraintsshadow mode, explicit validators, Simplex fallback
Privacy/security leakageOperational schedules or restricted zones are exposednetwork segmentation, least-privilege message routing, signed/encrypted links

Sources ​

Public research notes collected from public sources.

Copyright © 2026 kvynlim